> Re: 2.4.19-preX: What we really need: -AA patches finally in the tree
>
> From: Martin J. Bligh ([email protected])
> Date: Mon Feb 25 2002 - 20:32:06 EST
>
> > Not to begin the flamewar, but no thanks. rmap-12f blows -aa away AFAIK
> > on this P200 w/ 64MB ram.
>
> rmap still sucks on large systems though. I'd love to see rmap
> in the main kernel, but it needs to get the scalability fixed first.
> The main problem seems to be pagemap_lru_lock ... Rik & crew
> know about this problem, but let's give them some time to fix it
> before rmap gets put into mainline ....
>
> Martin.
>
Martin,
just out of curiosity: where does "large systems" start in your
context?
Martin
--
------------------------------------------------------------------
Martin Knoblauch | email: [email protected]
TeraPort GmbH | Phone: +49-89-510857-309
C+ITS | Fax: +49-89-510857-111
http://www.teraport.de | Mobile: +49-170-4904759
On Wed, 27 Feb 2002, Martin Knoblauch wrote:
> > > Not to begin the flamewar, but no thanks. rmap-12f blows -aa away AFAIK
> > > on this P200 w/ 64MB ram.
> >
> > rmap still sucks on large systems though. I'd love to see rmap
> > in the main kernel, but it needs to get the scalability fixed first.
> > The main problem seems to be pagemap_lru_lock ... Rik & crew
> > know about this problem, but let's give them some time to fix it
> > before rmap gets put into mainline ....
>
> just out of curiosity: where does "large systems" start in your
> context?
My guess it would start at about 4 or 8 CPUs.
Systems which have a lot of pagetable overhead would also
suffer with -rmap, until -rmap supports pte_highmem.
regards,
Rik
--
"Linux holds advantages over the single-vendor commercial OS"
-- Microsoft's "Competing with Linux" document
http://www.surriel.com/ http://distro.conectiva.com/
Rik van Riel wrote:
>
> On Wed, 27 Feb 2002, Martin Knoblauch wrote:
>
> > > > Not to begin the flamewar, but no thanks. rmap-12f blows -aa away AFAIK
> > > > on this P200 w/ 64MB ram.
> > >
> > > rmap still sucks on large systems though. I'd love to see rmap
> > > in the main kernel, but it needs to get the scalability fixed first.
> > > The main problem seems to be pagemap_lru_lock ... Rik & crew
> > > know about this problem, but let's give them some time to fix it
> > > before rmap gets put into mainline ....
> >
> > just out of curiosity: where does "large systems" start in your
> > context?
>
> My guess it would start at about 4 or 8 CPUs.
>
> Systems which have a lot of pagetable overhead would also
> suffer with -rmap, until -rmap supports pte_highmem.
>
So, what about a Dual-Athlon system with 2-3 GB of memory? Large
system, or just a peanut? :-)
Martin
--
------------------------------------------------------------------
Martin Knoblauch | email: [email protected]
TeraPort GmbH | Phone: +49-89-510857-309
C+ITS | Fax: +49-89-510857-111
http://www.teraport.de | Mobile: +49-170-4904759
On Wed, 27 Feb 2002, Martin Knoblauch wrote:
> Rik van Riel wrote:
> > On Wed, 27 Feb 2002, Martin Knoblauch wrote:
> >
> > > > > Not to begin the flamewar, but no thanks. rmap-12f blows -aa away AFAIK
> > > > > on this P200 w/ 64MB ram.
> > > >
> > > > rmap still sucks on large systems though. I'd love to see rmap
> > > > in the main kernel, but it needs to get the scalability fixed first.
> > > > The main problem seems to be pagemap_lru_lock ... Rik & crew
> > > > know about this problem, but let's give them some time to fix it
> > > > before rmap gets put into mainline ....
> > >
> > > just out of curiosity: where does "large systems" start in your
> > > context?
> >
> > My guess it would start at about 4 or 8 CPUs.
> >
> > Systems which have a lot of pagetable overhead would also
> > suffer with -rmap, until -rmap supports pte_highmem.
>
> So, what about a Dual-Athlon system with 2-3 GB of memory? Large
> system, or just a peanut? :-)
Unless you're overburdening the system with Oracle -rmap should
run just fine.
regards,
Rik
--
"Linux holds advantages over the single-vendor commercial OS"
-- Microsoft's "Competing with Linux" document
http://www.surriel.com/ http://distro.conectiva.com/
>> rmap still sucks on large systems though. I'd love to see rmap
>> in the main kernel, but it needs to get the scalability fixed first.
>> The main problem seems to be pagemap_lru_lock ... Rik & crew
>> know about this problem, but let's give them some time to fix it
>> before rmap gets put into mainline ....
>
> just out of curiosity: where does "large systems" start in your
> context?
I think the last tests I was doing was on a 12cpu box, which
gave me about a 40% performance degredation by installing rmap.
Most people won't be interested in anything near that large,
but as Rik said, the problems will start on much smaller systems.
One of the nice things about running really large systems is
that they show up scalability problems and race conditions
really quickly and obviously ;-)
If someone has data for 4 or 8 way systems, I'd love to see it.
Else I'll get some myself in a little while.
Thanks,
Martin.